Developing a Machine Learning Prediction Model for Cognitive Dysfunction of Community-Dwelling Older Adults

Abstract

<h3>Research Objectives</h3> To investigate predictors of cognitive dysfunction in older adults using machine learning algorithms. <h3>Design</h3> Cross-sectional study. <h3>Setting</h3> Data came from the 2018 Korean Longitudinal Study of Aging. <h3>Participants</h3> A total of 4,282 community-dwelling older adults. <h3>Interventions</h3> Not applicable. <h3>Main Outcome Measures</h3> We used Mini-Mental State Examination-K (MMSE-K) scores as the dependent variables (cognitive dysfunction). The independent variables included grip strength, the Center for Epidemiologic Studies Depression Scale (CES-D) scores (depression), life habits (e.g., smoking, drinking, presence of chronic disease, social activity participation), and demographic characteristics. Two machine learning algorithms, including random forest and classification and regression tree (CART), were compared to predict the cognitive dysfunction of older adults. A prediction model with lower mean absolute error (MAE) and root mean square error (RMSE) was selected for the data analysis. Partial dependence was analyzed to examine the effect of changes in important variables on the cognitive dysfunction of older adults. <h3>Results</h3> There were more female participants (n = 2,376, 55.5%) with the mean age of 75.7 years (SD = 8.1). We selected a random forest algorithm (random forest: MSE = 3.0, RMSE = 4.0, CART: MSE = 3.3, RMSE = 4.4). The random forest model identified depression, age, grip strength, educational attainment, self-reported health as predictors of cognitive dysfunction in older adults (in order of importance of the predictors). Also, the CES-D score showed a decline in the cognitive function above 4 points. It was noted that age was an important predictor of an increase in cognitive dysfunction (or decrease in cognitive function) only for older adults who are older than 65 years old (specific age). In addition, an increase in cognitive function was related to an increase in grip strength from 0 kg to 20 kg. <h3>Conclusions</h3> We presented important variables that predict older adults' cognitive dysfunction. Among various predictors, depression was the most contributing predictor to older adults' cognitive dysfunctions. These results suggest an association between cognitive dysfunction and depression in older adults. <h3>Author(s) Disclosures</h3> We have no financial disclosure or conflicts of interest.